Sophia:Network: Difference between revisions

From Grid5000
Jump to navigation Jump to search
No edit summary
Line 55: Line 55:
== High Performance networks ==
== High Performance networks ==


== Myri 10G ==
=== Myri 10G ===
A subset of the sol cluster (sol-[29-50]) is connected to a 10G Myrinet switch.
A subset of the sol cluster (sol-[29-50]) is connected to a 10G Myrinet switch.
=== Infiniband 40G on uva and uvb ===
uva and uvb cluster nodes are all connected to 40G infiniband switches. Since these two clusters are shared with the Nef procution cluster at INRIA Sophia, we are using Infiniband partitions to isolate the nodes from nef when they are available on grid5000. The partition dedicated to grid5000 is '''0x8108'''. The ipoib interfaces on nodes are therefore named ib0.8100 instead of ib0. (for openmpi you must use: <code>btl_openib_pkey = 0x8108 </code>)
==== Nodes ====
*<code class="host">uva-1</code> to <code class="host">uva-13</code> and <code class="host">uvb-1</code> to <code class="host">uvb-44/code>have one QDR Infiniband card.
* Card Model : Mellanox Technologies MT26428 [ConnectX IB QDR, PCIe 2.0 5GT/s].
* Driver : <code class="dir">mlx4_ib</code>
* OAR property : ib_rate=40
* IP over IB addressing :
** <code class="host">uva-[1..13]-ib0</code>.sophia.grid5000.fr ( 172.18.131.[1..13] )
** <code class="host">uvb-[1..44]-ib0</code>.sophia.grid5000.fr ( 172.18.132.[1..44] )
==== Switch ====
* three Mellanox IS50xx QDR switch Infiniband Switch
* Topology available here : https://wiki.inria.fr/ClustersSophia/Network (uva and uvb are nef084-nef140 on the nef production cluster)
==== Interconnection ====
Infiniband network is physically isolated from Ethernet networks. Therefore, Ethernet network emulated over Infiniband is isolated as well. There isn't any interconnexion, neither at the data link layer nor at the network layer.


==  Architecture ==
==  Architecture ==
[[Image:Sophia-cluster_modified.png|center|thumb|view of the Baie|600px]]
[[Image:Sophia-cluster_modified.png|center|thumb|view of the Baie|600px]]

Revision as of 08:58, 30 September 2016

Network Topology

IP networks in use

You have to use a public network range to run an experiment between several Grid5000 sites.

Public Networks

  • computing : 172.16.128.0/20
  • virtual : 10.164.0.0/14

Local Networks

  • admin or ipmi : 172.17.128.0.24

Network

Physical view of the compute network

Gigabit Ethernet

Sol Cluster (Sun X2200 M2)

All the nodes are connected (in a non blocking way) to the main Foundry FastIron Super X switch.

47 nodes (from sol-1 to sol-47) have a second Ethernet interface connected to 2 Cisco 3750 stacked switches but they are not activated.

  • IP addressing for eth1: sol-{1-47}-eth1, using the 172.16.128.0/24 network ( 172.16.128.154 can be used as a gateway)

The Cisco are interconnected to the main FastIron router with a single 1Gbps link.

Warning.png Warning

The second network is no longer available on sol, as one of the two Cisco switch is dead

Suno Cluster (Dell R410)

All the nodes are connected (in a non blocking way) to the main Foundry FastIron Super X switch.

Topology

The main switch is a Foundry FastIron Super X. It has 2 dual 10GB modules, 4 modules with 24 gigabit ports, and 12 gigabit ports on the management module (so 60 gigabit ports are available). 4 slots are currently free.

The EdgeIron stack (helios) is connected to the FastIron with 2 aggregated 10GB links.

Network usage is available through cricket at: https://helpdesk2.grid5000.fr/cricket/grapher.cgi?target=%2Fsophia%2Fgw-sophia.sophia.grid5000.fr

High Performance networks

Myri 10G

A subset of the sol cluster (sol-[29-50]) is connected to a 10G Myrinet switch.

Infiniband 40G on uva and uvb

uva and uvb cluster nodes are all connected to 40G infiniband switches. Since these two clusters are shared with the Nef procution cluster at INRIA Sophia, we are using Infiniband partitions to isolate the nodes from nef when they are available on grid5000. The partition dedicated to grid5000 is 0x8108. The ipoib interfaces on nodes are therefore named ib0.8100 instead of ib0. (for openmpi you must use: btl_openib_pkey = 0x8108 )

Nodes

  • uva-1 to uva-13 and uvb-1 to uvb-44/code>have one QDR Infiniband card.
  • Card Model : Mellanox Technologies MT26428 [ConnectX IB QDR, PCIe 2.0 5GT/s].
  • Driver : mlx4_ib
  • OAR property : ib_rate=40
  • IP over IB addressing :
    • uva-[1..13]-ib0.sophia.grid5000.fr ( 172.18.131.[1..13] )
    • uvb-[1..44]-ib0.sophia.grid5000.fr ( 172.18.132.[1..44] )

Switch

Interconnection

Infiniband network is physically isolated from Ethernet networks. Therefore, Ethernet network emulated over Infiniband is isolated as well. There isn't any interconnexion, neither at the data link layer nor at the network layer.


Architecture

view of the Baie